​Background: Artificial intelligence (AI) chatbots are increasingly consulted for dental aesthetics information. This study evaluated the performance of multiple large language models (LLMs) in answering patient questions about tooth whitening. Methods: 109 patient-derived questions, categorized into five clinical domains, were submitted to four LLMs: ChatGPT-4o, Google Gemini, DeepSeek R1, and DentalGPT. Two calibrated specialists evaluated responses for usefulness, quality (Global Quality Scale), reliability (CLEAR tool), and readability (Flesch-Kincaid Reading Ease, SMOG index). Results: The models generated consistently high-quality information. Most responses (68%) were "very useful" (mean score: 1.24±0.3). Quality (mean GQS: 3.9±2.0) and reliability (mean CLEAR: 22.5±2.4) were high, with no significant differences between models or domains (p>0.05). However, readability was a major limitation, with a mean FRE score of 36.3 ("difficult" level) and a SMOG index of 11.0, requiring a high school reading level. Conclusions: Contemporary LLMs provide useful and reliable information on tooth whitening but deliver it at a reading level incompatible with average patient health literacy. To be effective patient education adjuncts, future AI development must prioritize readability simplification alongside informational accuracy.​